Matched maneuver detector
A target tracking arrangement predicts the state of a target. The predictor may be a Kalman filter. In the presence of a target which is maneuvering, the prediction may be in error. A maneuver detector is coupled to receive residuals representing the difference between the predictions and the target state. The maneuver detector is matched to the predictor or Kalman filter to thereby tend to reduce the undesirable effects of system noise. The matching may be of the frequency response.
Latest Lockheed Martin Corporation Patents:
This invention relates to automatic target tracking, and more particularly to target maneuver detection in automatic tracking systems.
BACKGROUND OF THE INVENTIONAutomatic tracking systems are widely used for various military and nonmilitary purposes. Among the military purposes are the determination of the locations of airborne, undersea, and seaborne vehicles in a region to be controlled, and the automatic control of weapons directed toward such targets. Air traffic and harbor traffic control are among the civilian or peacetime uses of automatic tracking systems.
State information from sensor 12 of
Those skilled in the art know that predictors such as 28 of
A maneuver detector illustrated as a block 32 is coupled to receive the residual information from error detector 26 of
Target tracking with improved or alternative maneuver detection is desired.
SUMMARY OF THE INVENTIONA method for maneuver detection in a target tracking context includes the steps of generating data relating to the state of a target. This data may include position and velocity, velocity and acceleration, or other derivatives or integrals of position and rate of change of position. The method also includes the step of predicting the state of the target at a selected time by use of a predictor having a known response to an instantaneous change in velocity of the target. The state of the target at the selected time is compared with the predicted state, to thereby generate a residual. The residual is detected and filtered with a detector having a response which is ideally identical to, or which is tuned to, impulse (acceleration) response of the residual. In a particularly advantageous mode of the method, the known impulse response includes a low-pass characteristic and is optimal in a theoretical matched filter sense.
The invention is based upon the understanding that the frequency characteristics of the maneuver detector should match the impulse response of the predictor of the Kalman filter tracking system. A maneuver is declared when the value of the output of the matched-filter maneuver detector exceeds a given value or threshold.
In general, the assumption is made for purposes of determining the response of the Kalman filter that the target has been flying with a fixed velocity, and at some moment in time undergoes impulse acceleration. Thus, the target is assumed to change velocity instantaneously from the original fixed velocity to a new velocity. This corresponds to an infinite velocity slope, corresponding to an acceleration impulse. This is a convenient mathematical fiction which allows testing or modeling of the residual response of the Kalman filter. The modeling of the residual response characterizes the frequency response of the residual. The maneuver detector in the prior art looked for a bias in the residual. Noise in such prior-art systems can result in a non-zero value in the residual. In order to avoid false declarations of maneuvers, the maneuver detector must ignore such non-zero values caused by noise. Noise tends to have a higher frequency than a bias caused by true target acceleration. Thus, “low-pass filtering” of the residual in the frequency domain tends to reduce the relative magnitude of noise in the maneuver detector response. The impulse response of the residual of the Kalman filter identifies the maximum possible frequency associated with a true maneuvering target. Thus, matching of the frequency response of the maneuver detector to the impulse response of the residual of the Kalman filter makes the maneuver detector, in principle, sensitive to the residual frequencies associated with targets, but not with noise.
The procedure for determining the impulse response of the Kalman filter is illustrated in the flow diagram of
β=2(2−α)−4√{square root over (1−α)} (1)
α=−⅛(I2+8I−(I+4)√{square root over (I2)}+8I) (2)
where
I=t2σw/σm;
σw is the process noise uncertainty;
σm is the measurement uncertainty; and
T is the update interval.
The second step of the a priori determination of the coefficients of the predictor 24 of
The “truth” data is then effectively “run through” the Kalman filter to identify the residual response, as set forth by block 416 of
sk+1=xk+1,truth−{circumflex over (x)}k+1 (4)
where:
{tilde over (x)}k is the corrected (smoothed) position at time k;
{tilde over (v)}k is the corrected velocity at time k;
{circumflex over (x)}k+1 is the predicted position at time k+1;
{circumflex over (v)}k+1 is the predicted velocity at time k+1;
sk+1 is the residual value at time k+1;
xk+1,truth is the true target position at time k+1; and
α and β are the steady-state Kalman filter gains. Equations 3, 4, and 5 together simulate the ideal impulse response of the predictor to a unit acceleration impulse. The frequency response of the Kalman filter is implicit in the calculated result.
The response of the residual of the Kalman filter, illustrated as 314 in
where
yi=matched response; and
k_max is that value of j for which the tail of the response becomes insignificant, which is merely an implementation choice.
A method for maneuver detection in a target tracking context includes the steps of generating data (12, 18) relating to the state (16) of a target (T). This data (12, 18) may include position and velocity (310, 312), velocity and acceleration, or other derivatives or integrals of position and rate of change of position. The method also includes the step of predicting the state of the target at a selected time by use of a predictor (28) having a known response (314) to an instantaneous change in velocity (310, t1, 312) of the target (T). The predictor may be part of a Kalman filter. The state (16) of the target (T) at the selected time is compared with the predicted state, to thereby generate a residual. The residual is detected with a detector (32) having a response, such as a frequency response, identical to the known response (314). In a particularly advantageous mode of the method, the known response includes low-pass frequency characteristics (510).
Claims
1. A method for maneuver detection in target tracking, comprising the steps of:
- generating data relating to the state of a target;
- predicting from said data the state of said target at a selected time by use of a Kalman predictor having a known frequency response to an instantaneous change in velocity of said target;
- comparing the state of said target at said selected time with said predicted state, to thereby generate a Kalman residual; and
- detecting said Kalman residual with a detector having a frequency response identical to that of said known frequency response.
2. A method according to claim 1, wherein said known frequency response includes low-pass characteristics.
3. A method according to claim 1, wherein said step of generating data relating to the state of a target includes the step of generating data relating to at least one of (a) position and velocity and (b) velocity and acceleration of said target.
4. A method according to claim 1, wherein said step of predicting the state of said target at a selected time by use of a Kalman predictor having a known frequency response to an instantaneous change in velocity of said target includes the step of taking the difference between said data relating to the state of a target and a predicted value.
5. A method according to claim 1, wherein said step of detecting includes the steps of generating a Kalman residual and applying said Kalman residual to a threshold.
6. A method for maneuver detection in target tracking, comprising the steps of:
- generating data relating to velocity state of a target;
- predicting the state of said target from said data by use of a Kalman predictor having a known frequency response to two mutually offsetting instantaneous changes in velocity of said target;
- comparing the state of said target at said selected time with said predicted state, to thereby generate a Kalman residual; and
- detecting said Kalman residual with a detector having a frequency response identical to that of said known response.
7. A method according to claim 6, wherein said step of predicting the state of said target from said data by use of a Kalman predictor having a known frequency response to two mutually offsetting instantaneous changes in velocity of said target includes the step of using a Kalman predictor having a known frequency response to an impulse acceleration.
3135053 | June 1964 | Newman et al. |
3646588 | February 1972 | Van Popta |
3869601 | March 1975 | Metcalf |
4038536 | July 26, 1977 | Feintuch |
4038657 | July 26, 1977 | Orloff |
4128837 | December 5, 1978 | Page |
4179696 | December 18, 1979 | Quesinberry et al. |
5291199 | March 1, 1994 | Overman et al. |
5422829 | June 6, 1995 | Pollock |
5428562 | June 27, 1995 | Gay |
5451960 | September 19, 1995 | Kastella et al. |
5533067 | July 2, 1996 | Jamal et al. |
5537118 | July 16, 1996 | Appriou |
5604683 | February 18, 1997 | Roecker |
5631653 | May 20, 1997 | Reedy |
5638281 | June 10, 1997 | Wang |
5727032 | March 10, 1998 | Jamal et al. |
5751777 | May 12, 1998 | Zampetti |
5801970 | September 1, 1998 | Rowland et al. |
5810014 | September 22, 1998 | Davis et al. |
5921937 | July 13, 1999 | Davis et al. |
5943381 | August 24, 1999 | Zampetti |
5995620 | November 30, 1999 | Wigren |
6064333 | May 16, 2000 | Stromberg |
6243037 | June 5, 2001 | Pulford et al. |
6260759 | July 17, 2001 | Nguyen et al. |
6262680 | July 17, 2001 | Muto |
6377682 | April 23, 2002 | Benesty et al. |
6580384 | June 17, 2003 | Fung et al. |
6683968 | January 27, 2004 | Pavlovic et al. |
6710743 | March 23, 2004 | Benner et al. |
6870793 | March 22, 2005 | Ishihara et al. |
6873903 | March 29, 2005 | Baiada et al. |
6999601 | February 14, 2006 | Pavlovic et al. |
20020187765 | December 12, 2002 | Kozak et al. |
20030118246 | June 26, 2003 | August |
20030156762 | August 21, 2003 | August |
20030202426 | October 30, 2003 | Ishihara et al. |
0323688 | July 1989 | EP |
Type: Grant
Filed: Aug 27, 2004
Date of Patent: Mar 6, 2007
Assignee: Lockheed Martin Corporation (Bethesda, MD)
Inventor: Robert E. Yang (Cherry Hill, NJ)
Primary Examiner: Bernarr E. Gregory
Attorney: Duane Morris, LLP
Application Number: 10/928,853
International Classification: G01S 13/66 (20060101); G01S 13/06 (20060101); G01S 15/06 (20060101); G01S 15/66 (20060101); G01S 17/06 (20060101); G01S 17/66 (20060101); G01S 13/00 (20060101); G01S 15/00 (20060101); G01S 17/00 (20060101);